I’m reshaping this account around the OSS/dev tooling work I care about most
I’ll be sharing reproducible workstations, AI-native developer workflows, app scaffolding, DevEx, persistent context for coding agents, and V scientific tooling
Most of it already lives on my GitHub
🚀 VSL v0.2.0-beta.1 and VTL v0.2.0-beta.1 are out!
🔬 VSL = scientific computing foundation
🧠 VTL = tensors, autograd, and neural networks
One step closer to real-world ML in V
https://t.co/gaRJCxAzWF
https://t.co/6SBdLYtYiC
@KaiXCreator Exactly.
That’s one of the main reasons I started building my own AI workflow around reproducible workstations, persistent context, repo conventions, skills, and local tooling.
If everyone has access to the same models, the differentiator becomes the system around them.
@CodeWithAmann Arch Linux daily here.
My setup has slowly turned into a full reproducible workstation with chezmoi, Hyprland/Wayland, theme-aware colors, and a lot of automation around my dev workflow.
I keep the dotfiles open here: https://t.co/mCx4BO1AyY
@ThePrimeagen This is why I care more about the harness around the agent than the agent itself.
Models and apps will keep changing.
But repo context, memory, scripts, conventions, workflows, and guardrails need to be portable instead of locked into one tool.
@theo Yes. I still prefer CLI-first agents.
The terminal keeps the agent close to the repo, scripts, git, logs, tests, and my actual dev environment.
OpenCode and GitHub Copilot CLI have worked really well for me.
Desktop apps feel a bit too far from the loop.
@midudev Es una locura total!
Más allá de si técnicamente termina saliendo bien, este tipo de cambios me dejan pensando bastante en reviewability, ownership y mantenibilidad a largo plazo.
No me preocupa tanto “si compila”, sino cómo se razona y se evoluciona algo así después.
@hugorcd Same feeling.
I’m not against compiler-assisted rewrites or AI-assisted work, but massive diffs like this make me wonder how we’re supposed to reason about ownership, reviewability, and long-term maintainability.
The code may work today, but the process is what worries me.
@ThePrimeagen I feel this.
The more I use agents, the more I think the real work is not “make it code for me”, but “make the context, constraints, and feedback loop explicit enough that it can work with me”.
i never thought i'd say this, but coding got boring
you write a prompt, wait 10-50 minutes, review the code, ask for fixes, then repeat
that's it. that's the job now
no more debugging for hours. no more stack overflow rabbit holes
just… prompting. reviewing. prompting again
VSL v0.1.51 released!
Major milestone: VSL now has 100% pure V BLAS and LAPACK implementations as the default!
- Zero deps (no C libs needed)
- High perf (competitive with C backends)
- Cross-platform
- Production ready
- Benchmarks and enhanced docs
https://t.co/90jPDKLubk
"Users don't care about performance"
They do. They just rarely have a choice. Give them to choose between a slow and a fast app and you will be blown away by the results. The problem is that all apps these days are a slow, buggy crap. It's you who don't care. Not them.
Just updated my GitHub profile! Check it out, maybe you'll find something useful in there! 👋🔭 Will try to keep it up to date every month 😁
https://t.co/oESGRbRgPk
V 0.4 is out!
2160 commits pushed to master, 1205 bugs closed since V 0.3.
4 165 files changed, 329 174 insertions(+), 42 434 deletions(-).
V 0.4 changelog is combined from 0.3.1-0.3.5 releases.
The changelog is huge! It has 370 items:
https://t.co/wxCnmW7ymB